{"paper":{"title":"DriveSafer: End-to-End Autonomous Driving with Safety Guidance","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A safety framework for end-to-end driving planners cuts catastrophic failures by 48 percent on the NAVSIM benchmark.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Raj Rajkumar, Shounak Sural","submitted_at":"2026-05-16T01:21:30Z","abstract_excerpt":"End-to-End (E2E) autonomous driving models have shown growing capability in recent years, with performance improving on increasingly challenging benchmarks. However, modern generative E2E planners still suffer from a substantial number of catastrophic failures in safety-critical scenarios. We find that many such failures arise from violations of physical constraints and safety requirements, leading to unsafe behavior. Motivated by this finding, in this paper, we focus on improving safety outcomes in generative end-to-end driving with a targeted reduction of catastrophic planning failures, inst"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Compared to the state-of-the-art DiffusionDrive model, on the NAVSIM benchmark, DriveSafer reduces the number of catastrophic failures (PDMS=0) by 48%, with over 65% reduction in drivable-area compliance failures.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The claim that many catastrophic failures arise specifically from violations of physical constraints and safety requirements (abstract, paragraph beginning 'We find that many such failures arise...') and that adding training-time constraints plus inference-time guidance will reduce those failures without creating new failure modes or degrading performance on non-catastrophic cases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DriveSafer reduces catastrophic failures (PDMS=0) by 48% and drivable-area compliance failures by over 65% versus DiffusionDrive on the NAVSIM benchmark by combining training-time safety constraints with inference-time guidance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A safety framework for end-to-end driving planners cuts catastrophic failures by 48 percent on the NAVSIM benchmark.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b93d99b4d36fad0a98c27ab7016ef5e2563ead07793ad535a026a79241187b35"},"source":{"id":"2605.16737","kind":"arxiv","version":1},"verdict":{"id":"3c37678d-7c04-460c-9821-0113dbd00709","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:39:31.016009Z","strongest_claim":"Compared to the state-of-the-art DiffusionDrive model, on the NAVSIM benchmark, DriveSafer reduces the number of catastrophic failures (PDMS=0) by 48%, with over 65% reduction in drivable-area compliance failures.","one_line_summary":"DriveSafer reduces catastrophic failures (PDMS=0) by 48% and drivable-area compliance failures by over 65% versus DiffusionDrive on the NAVSIM benchmark by combining training-time safety constraints with inference-time guidance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The claim that many catastrophic failures arise specifically from violations of physical constraints and safety requirements (abstract, paragraph beginning 'We find that many such failures arise...') and that adding training-time constraints plus inference-time guidance will reduce those failures without creating new failure modes or degrading performance on non-catastrophic cases.","pith_extraction_headline":"A safety framework for end-to-end driving planners cuts catastrophic failures by 48 percent on the NAVSIM benchmark."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16737/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:01:19.893071Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:51:05.149926Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.337383Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.465896Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"1702686316eff1dfa5d190f535dbf7bd33c1f440042663e8df5454dc949ae5f0"},"references":{"count":44,"sample":[{"doi":"","year":2024,"title":"Real-time mpc with control barrier functions for autonomous driving using safety enhanced col- location.IFAC-PapersOnLine, 58(18):392–399, 2024","work_id":"20fb2902-6df3-4465-a053-1cae08a8ff8a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"NuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles","work_id":"b16aace6-deff-4546-b333-bcb7c9c07cdb","ref_index":2,"cited_arxiv_id":"2106.11810","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:2506.04218 (2025)","work_id":"d8bb0a5f-f4f1-4d7d-a9b1-90eacc293742","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Learning by cheating","work_id":"9a2433c5-4f0f-42d5-84d3-1ee78cdf5336","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"VADv2: End-to-End Vectorized Autonomous Driving via Probabilistic Planning","work_id":"e7670f83-e1e1-41e7-86eb-39477a3a10b2","ref_index":5,"cited_arxiv_id":"2402.13243","is_internal_anchor":true}],"resolved_work":44,"snapshot_sha256":"ea8da5ab5f7be0c1be7e006b9bab2db1bcf70f218e5922fd4e7e0e659218291b","internal_anchors":8},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0cf765e1f48649da7f9b7510de0e8b5af0625f447844136f048c73c3d7f4beaf"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}